- From: Teoh Teik Huat <peter@iti.gov.sg>
- Date: Mon, 4 Mar 1996 14:15:02 +0800 (SST)
- To: harold@driscoll.chi.il.us (Harold A. Driscoll)
- Cc: www-html@w3.org
> > I agree--what you suggest is very much needed. The "URN" proposals are an > attempt to accomplish this. By way of analogy, we refer to a book via an > ISBN, which identifies the book, rather than the store and shelf which hosts it. > > For example, if I want to get a copy of RFC 1866, I don't care were it comes > from, only that it arrives efficiently and is a legitimate copy. The > InterNIC in this case has quite a few mirrors around the world, and lots of > others have copies as well. > > I've attempted to follow this development, but am not as knowledgable as I > might like. My notes at http://www.ccs.org/validate/wwwspec5.html may be > helpful to you. > > Good luck. /Harold > > At 13:54 3/3/96 +0800, Teoh Teik Huat wrote: > >> > >> Don't redirects solve that? I mean, if you're running > >> a spider (I run MOMspider) you'll see the redirects > >> and update the links. > >> > >> I thought everyone was doing this. What I do > >> is run MOMspider once a week, and update > >> redirected links to their new location. If I change > >> the location of my pages, I usually re-direct for a while, > >> then stop (since everyone else who was running a > >> spider has noticed that the location has changed). > >> > >> Also, a lot of servers have virtual paths, so you > >> could be doing the same thing with that. > >> > >> >>> Bryan A. Bentz <bentz@martigny.ai.mit.edu> - 3/2/96 10:26 AM >>> > >> It often seems that URL's point to (moved) locations... it kind of > >> reminds me of the garbage collection problem in Lisp, in which data > >> can be moved around in memory and all active pointers must be > >> updated. > >> > >> One solution (not a good one) would be to treat it the same way, and > >> have a > >> GC web walker which re-points moved URL's; obviously this is fraught > >> with problems. > >> > >> However, an indirect referencing scheme might go a long way towards > >> solving this problem. Rather than: > >> > >> <A HREF="http://www.blah/foo.html"> which is an "absolute" > >> pointer, perhaps a syntax change allowing dynamic lookup would solve > >> the problem: > >> <A HREF="www.blah.foo.html@index-server.com" ...> > >> > >> The idea is that to resolve the HREF, (in this case) index-server.com > >> is asked for the *current* location of www.blah.foo.html (syntax here > >> needs some thought). This way, one could move pages around, update > >> pointers on whatever index server one uses, and all existing pointers > >> would still work. > >> > > > >My idea may be a stupid idea, but I would like to learn that > >it is stupid. > > > >To increase reliability and balance the load of WWW servers, in > >addition to having indirect links (dynamic query of servers for > >links), we can have multiple links - where web pages are duplicated > >and distributed in a few sites. The resolution of which path > >(or servers) to go to for WEB pages therefore depends on the > >path's availability and load condition. If a few simultaneous calls > >comes in, they can be redirected to different sites for WEB pages. > > May be I should include an aritlce from the Internic site - 96mar directory, which emphasize the importance of performance in protocol specification: >Group Name: Applications/Transport Joint Session (apptsv) >Chairs: Harald Alvestrand <harald.t.alvestrand@uninett.no> > Allison Mankin <mankin@isi.edu> >=================================================== >TRANSPORT/APPS JOINT BOF ON WEB-RELATED TRANSPORT ISSUES >-------------------------------------------------------- > >This session will attempt to address the problems that the phenomenal >growth of the Web has caused, both related to the strains it puts on >the Web servers and the strain it puts on the infrastructure. > >The session will start with some short introductions to illuminate >various parts of the problem and some examples of proposed solutions, >and thereafter, there will be the usual open discussion. > >Examples of the problems we want to address: > >- - Web problems for the Web client > - "always" goes far away > - Takes too long >- - Web problems for the Web servers > - Thousands of connections in CLOSE_WAIT and similar states > - Hundreds of addresses per interface > - Nasty behaviour by clients (they crash) >- - Web problems for the network > - Congestion because TCP backoff doesn't have time to establish > - Routing cache miss because of too many destinations at once > - Loss of "locality of reference" model > >Examples of solutions we might want to pursue: > >- - An extensive caching network might solve the problem. >- - Deploying T/TCP might solve the problem. >- - Deploying LBL reliable multicast in unicast mode rather than TCP > might solve the problem. >- - Switching to HTTP/NG with multiple-stream persistent connections > might solve the problem. >- - Deploying flow policing in the routers rather than the end nodes > might solve the problem. > >Some of these will be presented in short introductions. > >After the presentations, there will be a general discussion, >terminating in an attempt to decide on what further work, if any, we >need to do related to this problem area. Peter Teoh Information Technology Institute Internet : peter@iti.gov.sg Science Park II Tel : 65-7705585 11 Science Park Road Fax : 65-7791827 Singapore 117685
Received on Monday, 4 March 1996 01:09:55 UTC